Mainboard VRM layouts

[BotDamian]

Hi everyone,

I like to learn about things and I know that there is a PWM signal that is passed to mosfets but are there other variants of this?

Some mainboards have highside and lowside mosfets like my B450 Mortar Max and how does that work?
Does it go:
pwm ctrl -> highside mosfet -> 2x lowside mosfet? but this would mean it has doublers or how? this is what I don't understand.

 

## Edit ##

Calculating Mosfet/VRM capabilities

 

VRM phases/layouts

 

[Morgan MLGman]

There's no better person to explain this than Buildzoid himself, I believe you'll learn the most if you watch this:

 

2 minutes ago, BotDamian said:

but this would mean it has doublers or how? this is what I don't understand

Doublers are for something different, AFAIK they split the load of each phase onto two identical sets of VRM components.

[Haro]

No, doublers split the pwm signal and and split it for 2 phases, some do current monitoring and extend the pwm signal to one phase to current balance and that results in better efficiency. (like the ISL6617)

discrete mosfets (high side and low side) both do voltage regulating, it's just afaik the low side does a lot of the work, and the high side doesn't handle a lot of current. 

for example, the mortar/titanium (though i think they're the same), which iirc uses 4 SM4337 and   4 SM3503, does not use doublers.

 

edit: just wanted to say, if you want to know more about vrm, watch build zoid's pcb breakdowns.

Edited August 24, 2020 by TofuHaroto

[mariushm]

Doublers take one of the PWM signals from the VRM controller (one PWM signal per phase) and double it, producing TWO PWM signals - often if not almost always opposite to each other (when one pwm signal says turn on, the other pwm signal says turn off)

So if VRM says 1011  (turn phases 1 , 3, and 4 on and turn off phase 2) for the duration of the on / off period (usually 1 / 600 000 or 1 / 300 000 seconds) they'll do this : 

10 00 10 10   (for half the duration)

01 00 01 01  (for half the duration)

so overall over the duration of that 1/300k or 1/600k you have 11 00 11 11 but not at same time.

 

This way, they can use a cheaper or more mass produced vrm which let's say can only do 4 phases, but use doublers to have 8 phases, spreading the current over more phases / chips and reducing the heat produced per set of components and making the output voltage a bit smoother

 

A phase needs a hi-side mosfet , a lo-side mosfet and an inductor. 

hi-side takes in 12v at lower current so it can be a lower rating mosfet, which poorer characteristics, because it's stays on for shorter periods of time. 

lo-side outputs the low voltage (0.5v - 1.5v) but high current, so it's good idea to be very efficient, low resistance, very good performance mosfet. 

 

To keep costs down and/or to spread the heat (caused by the low efficiency of the mosfets) manufacturers sometimes parallel two lo-side mosfets (and sometimes two hi-side mosfets), because this way each lo-side mosfet gets half the current going through them, and each lo-side mosfet produces less heat, and the heat on the two lo-side mosfets can more easily be transferred into a heatsink because there's more surface touching the heatsink. 

 

Higher end VRMs use power stages which are basically chips that contain the driver, the hi-side and lo-side mosfets, in a package that's more efficient and which can be more easily cooled using heatsinks. Power stages also add smarts to the package, various monitoring functions, protections against too much current, too high heat, tricks to be more efficient during low power consumption and lots of other things. 

[BotDamian]

So it looks something like this?

If i'm not mistaken there are also twins right? So the 1 signal goes straight to two mosfets.

 

[Haro]

10 hours ago, BotDamian said:

So it looks something like this?

Kind of like that yes, doublers don't affect voltage regulation, they just split the pwm signal, while discrete mosfets (both the high side and the low side) do voltage regulation.

[BotDamian]

Just now, TofuHaroto said:

Kind of like that yes, doublers don't affect voltage regulation, they just split the pwm signal, while discrete mosfets (both the high side and the low side) do voltage regulation.

So the hiFET loFET is better than the normal PWM -> Mosfet phase?

[mariushm]

Think of the hi-side and lo-side mosfets as two valves, faucets, on/off switches.  

 

The mosfet driver turns on the hi-side mosfet for a very brief period of time and lets 12v come in and top up that phase with energy and then quickly turns it off.  Imagine a car motor and the motor spraying a bit of fuel in the cylinder if that's a better analogy. 

Then the mosfet driver turns on the lo-side mosfet and lets a bit of energy flow out into the inductor and capacitors and as soon as they do their job and the voltage on the capacitors raises to desired level (ex 1.3v if the processor wants 1.3v) the mosfet driver turns off the lo-side mosfet and repeats the whole cycle. 

The inductor is there to charge up with some energy and continue to provide energy while the hi-side mosfet is turned off. 

 

Look at those buildzoid videos they explain it quite well. 

 

[Haro]

10 hours ago, BotDamian said:

So the hiFET loFET is better than the normal PWM -> Mosfet phase?

No, mosfets integrate the high side and the low side and the driver IC, they produce less heat and are easier to cool than discrete (but it of course depends on the specific model), there are DRmos and SPS, SPS (smart power stages) integrate current monitoring and temp monitoring and some safety features, and dr mos don't. 

example for a DRmos, SIC639's or the 634's. 

example for SPS's, ISL99370's/60's/90's, 21472/21462's. 

again though, watch buildzoid, he does explain things quite well.

[mariushm]

3 minutes ago, BotDamian said:

So the hiFET loFET is better than the normal PWM -> Mosfet phase?

There's always a hi-fet and a lo-fet , or there's a "power stage" chip which integrates both mosfets in a single chip, sometimes along with the driver that controls the two mosfets. 

Some VRM controller chips have built in drivers, most don't, some only have the pwm signals which must go to a mosfet drivers which in turn control the mosfets. 

Some VRM controller chips have only 1 or 2 built-in drivers - there were motherboards which used such vrm controller chips with 6 phases, but arranged in 4 phases (4 pwm signals) for the cpu core which may go to doublers or mosfet drivers or directly to power stages (which contain driver)  and the 2 other phases may be used for the SoC or integrated graphics part, connecting hi-side and lo-side mosfets directly to the 2 built-in drivers. 

[BotDamian]

what is exactly soc? It's system on chip, but is it also the ram controller and so on is a good SOC vrm even important or only important for APU systems?

[Haro]

10 hours ago, BotDamian said:

what is exactly soc?

the soc is basically the imc, igpu (if it's the apu) and the interconnect essentially.

also quote people so they can see your reply.

[BotDamian]

Is this correct or have I done it wrong?

[Haro]

10 hours ago, BotDamian said:

Is this correct or have I done it wrong?

-snip-

the output filter depends, and there could be 2 inductors, but if we're talking about the mosfets, yea kind of like that.

[mariushm]

For future reference , you could always look up datasheets of various vrm controller chips and dc-dc switching controllers.

 

As an example, here's a 2 phase controller with built in drivers : NCP5383 - ONSMS14374-1.pdf

 

[BotDamian]

Is this the max switching frequency of a mosfet?

https://www.onsemi.com/pub/Collateral/NTTFS4C10N-D.PDF

[mariushm]

A mosfet doesn't have a fixed frequency.  It will sort of have a peak frequency based on the gate capacitance and resistance, how fast the gate can charge and open the connection, or the opposite discharge and close.

 

Typically a VRM controller on motherboards will work up to 600 kHz, most often when there's doublers involved, because that way doubler takes 600kHz pwm signal and you get 2 x 300 kHz. Where there's no doublers, the frequencies will be lower, 300-500 kHz.

 

Higher frequency is not always good, you get smaller components with higher frequency (smaller inductors, capacitors) but you may get less efficiency and in some cases more noise so more filtering would be needed. You also need stronger drivers to push that energy into the gate of mosfets and quickly charge up/discharge that gate capacitance to turn the mosfets really fast on and off.

 

That mosfet you linked to is not a particularly great mosfet for motherboard VRMs, the Rds(on) resistance is a bit high at 7-11mOhm, but i suspect due to package it's not aimed at that market

Figures 7-10 on page 5 looks like something that would help someone determine an optimal frequency based on the gate voltage and driver someone chooses.

[BotDamian]

19 hours ago, mariushm said:

A

When choosing a mainboard, how should I compare the VRM? Based on the current it can handle or the vrm features?

For example, csd87350q5d has a switching freq of max 1500khz and 40A, the isl9927b can handle 60A and is a SPS (smart power stage), but the  sm4503 can handle up to 80A.

 

So how do I compare VRMs? based on mosfets and the layout?

[mariushm]

You can't compare VRMs just by some parameters extracted out of context.

 

The peak frequency is meaningless as most VRMs will work in the 300-600 kHz range. Efficiency of a mosfet/power stage also varies with frequency, usually goes down as you increase frequency (you're trading a bit of efficiency for ability to use smaller components like inductors and output capacitors)

 

The peak current is meaningless on its own, because a VRM uses multiple phases, so the overall current consumed by a device is spread across all phases. For example, a CPU may consume 120w by receiving 1.2v at 100A, that 100A amount is spread across 8 mosfets/power stages so each mosfet/power stage handles 12.5A , which is well below 40A or 60A or 80A

 

The peak current is also deceiving, as in for example sm4503 *may* handle up to 80A, but it may need water cooling or lots of circuit board surface around it and a massive copper heatsink to dissipate all the heat produced, which is not realistic in a lot of cases (on a motherboard, you don't have an inch of circuit board that you can leave unpopulated and acting as heatsink)

 

An important property of each mosfet/power stage is Rds(on), how much resistance does the mosfet have when current flows through it. A higher Rds(on) means less efficiency, more heat is produced inside the chip, which has to be dissipated.

 

You can see in the datasheets of those parts

For example http://www.hardwaresecrets.com/datasheets/csd87350q5d.pdf

First page shows you the thing can do 40A, but it will produce nearly 6w of heat at that amount, it's most efficient at around 20A, staying under 2.5w :

 

 

So let's say you make a vrm with 8 of these and your cpu needs 120 watts (1.2v x 100A)... each of them the vrm will get 12.5A and dissipate around 1w of heat, so you get 8-10w of heat in the mosfets alone.

 

Now if you go to https://www.renesas.com/eu/en/doc/datasheet/isl99227-27b.pdf

go to page 10 and you can see there, for the same 6w of heat produced, the power stage can do up to around 50A (depends on output voltage, it's a bit more than 40A at 1.35v, the graph above is for 1.3v out) and has room for more if there's need.

At 12.5A, it will probably produce a bit less heat:

 

So a motherboard manufacturer may only need to use 6 of these instead of 8, because it knows if the cpu suddenly consumes 200A of current, there's margins.

 

Mobo maker also makes choices like 1$ heatsink vs 5$ heatsink ... sometimes 2$ more on heatsink saves you 3$ on mosfets. Sometimes a big heatsink is needed either way because marketing, people equate big heatsink with good vrm which is not always the case, so mobo needs one to sell well...

So as you can see it depends on lots of things.

[BotDamian]

18 hours ago, mariushm said:

So as you can see it depends on lots of things.

So the efficiency and head output is important I see,

but what if the heat output and efficiency is in the "golden range" for example 93% efficiency and has a good cooling.

What would be the next thing that would improve the voltage quality or improve the VRM? I've heard of doublers that have load balancing and can measure the load between mosfets, if one mosfet has a higher current flow, it will change the signal to distribute the load better between both mosfets.

 

I guess doublers are worse than real phases, but are good doubled mosfets (4*2) better than avg real phases (8)?

 

Same with good doublers that can load balance, measure temp, load etc. vs real phases.